From 5c152cbaa80ba1366b70ada95a88ba145f7e3d17 Mon Sep 17 00:00:00 2001 From: "github-actions[bot]" Date: Tue, 4 Aug 2026 06:38:48 +0000 Subject: [PATCH] [Doc][Misc] Add limitation for reduce sample Cherry-pick of PR #13468 by @zouzy5137 --- docs/source/user_guide/configuration/additional_config.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/source/user_guide/configuration/additional_config.md b/docs/source/user_guide/configuration/additional_config.md index e6a081fc3981..e4be575ab858 100644 --- a/docs/source/user_guide/configuration/additional_config.md +++ b/docs/source/user_guide/configuration/additional_config.md @@ -105,7 +105,7 @@ The following table lists additional configuration options available in vLLM Asc | `enable_dsa_cp` | bool | `False` | Whether to enable dsa_cp for DeepSeek V3.2, DeepSeek V4, and other models with the same architecture. This feature depends on FlashComm1. Please ensure that FlashComm1 is enabled before enabling this feature.| | `rejection_sampler_config` | dict | `{}` | Configuration options for rejection sampler (block verify and entropy verify). | | `multistream_dsv4_dsa_overlap` | bool | `True` | Whether to enable dsa multi-stream overlap for DeepSeek V4. | -| `enable_reduce_sample` | bool | `False` | Whether to enable reduce sample optimization to reduce communication and computation overheads in the tensor parallelism scenario. When enabled, logits are kept partitioned across TP ranks and only the small set of top-k candidate values/indices is communicated, instead of performing a full-vocabulary all-to-all/all-gather. | +| `enable_reduce_sample` | bool | `False` | Whether to enable reduce sample optimization to reduce communication and computation overheads in the tensor parallelism scenario. When enabled, logits are kept partitioned across TP ranks and only the small set of top-k candidate values/indices is communicated, instead of performing a full-vocabulary all-to-all/all-gather. **Limitations**: (1) PD disaggregation scenarios are not supported. (2) Must be disabled when sampling logprobs are requested. When reduce sample is enabled, logprobs are silently computed over partitioned logits instead of the full vocabulary, producing incorrect logprob values and top-k rankings.| The details of each configuration option are as follows: